An Ensemble Approach to Recognize Activities in Smart Environment Using Motion Sensors and Air Quality Sensors

Abstract

Listen

Translate article

AbstractSmart environments are built to capture Activities of Daily Living (ADL) data using a variety of sensors. Simple and sophisticated sensors sense the minute and intricate movements of an individual within the environment. With the growth in technology, applications of sensor data are extended to activity recognition. Motion sensors are able to detect movements of the person performing daily routine tasks such as cooking, cleaning, and eating. The utility of air quality data extracted from gas sensors for activity recognition is uniquely explored. The proposed model is built to identify activities using motion sensors and air quality forms of data individually. In this paper, the experiment focuses on routine tasks classification using pressure, temperature, and other related sensor data. Feature selection is performed on the air quality data using ANOVA-F Classification. The novel data is applied for activity recognition in a set of scenarios while simultaneously detecting the presence or absence of chemicals in the surrounding area. Machine learning techniques are applied to recognize activities performed by individuals in both smart environments. A comparative study is performed on the models generated for activity recognition. An ensemble form of learning, the Random Forest technique, provides the best prediction accuracy of 86.8% for activity prediction using motion sensor data. In the case of air quality data, the activity classification model produced an accuracy of 96.19% with the least feature set combination. Activity recognition helps in identifying changes in behavioral patterns and can be extended to assisted care benefiting healthcare professionals.KeywordsActivities of Daily LivingAir quality sensorsMotion sensorsHealth careMachine learning